KR101971875B1 - Device and method for diagnosing/predicting status of solar cell or solar cell array on real time based on machine learing - Google Patents

Abstract

According to an embodiment of the present invention, a machine learning based real-time photovoltaic cell or array state predicting method includes: (a) calculating an internal factor, an external factor, and performance data for a plurality of solar arrays each comprising a plurality of solar cells Collecting; (b) machine learning is performed through a predetermined machine learning model to define the relationship between the independent variable and the dependent variable by setting the internal factor and external factor as independent variables and setting the performance data as dependent variables ; (c) calculating predictive performance data for the solar array based on the machine learning, when there are external factors and internal factors collected in real time; And (d) providing the calculated predictive performance data through a user interface of an administrator terminal, wherein the internal factor includes a value for at least one of an internal temperature, a current, a voltage, and a power amount of the solar array Wherein the outlier comprises a value for at least one of an ambient temperature, an irradiation dose, a wind speed, a rainfall, a humidity and a fine dust concentration of the solar array, and the performance data includes a normal operation of the solar cell or the solar array And the like.

Description

TECHNICAL FIELD [0001] The present invention relates to a system and method for real-time photovoltaic cell or array status diagnosis / prediction based on machine learning,

The present invention relates to an apparatus and method for diagnosing / predicting a state of a real-time solar array based on machine learning.

In recent years, renewable energy systems are increasing in popularity as energy problems such as energy depletion and imbalance in energy supply and demand become serious with environmental problems such as climate change and environmental pollution caused by fossil energy use. Among them, the photovoltaic power generation system occupies the largest portion of various renewable energy systems due to the convenience of installation and relatively free installation conditions compared to other systems.

It is reported that the performance of the photovoltaic generation system that converts solar light energy into electric energy is influenced by changes in weather and environmental conditions such as solar radiation, outdoor temperature, module temperature, wind speed, and the like. However, although many research papers have been published on predictive models for the individual components that affect performance (ie, the temperature of the solar module), there are some variations in module temperature and weather and environmental conditions for the solar cell / , The prediction model based on complex influencing factors has not been developed.

As a result, there is no way to confirm the state of the solar cell every time, so the state of the cell depends on only the solar array on / off signal, so that it is not possible to cope with the inconvenience until a considerable period has elapsed There is no solution.

If only one cell in the array becomes defective, the whole cell becomes bad or malicious in the column. Therefore, the repair cost (replacement cost) and maintenance cost (operating cost) increase cumulatively to 15-20% have. This cell defect is very difficult to detect in advance and there is no system that can measure it in real time.

Although it is possible to find an alternative by attaching an anomaly detection sensor to each cell or monitoring the signal status by the inverter, it is expensive to install such a sensor or monitoring part in each of a plurality of arrays, It is not a fundamental solution because it requires re-maintenance.

In addition, the existing solar power plant monitoring system frequently makes the forecast of the power plant production deviations, and there are many problems that can not be dealt with in advance due to the inability to carry out analysis / prediction of facility cause early. In addition, the user interface of the existing system is not provided in a form that is easy for the administrator to grasp the system status at a glance, thus providing many inconveniences.

SUMMARY OF THE INVENTION The present invention provides a system and method for diagnosing / predicting a state of a real-time photovoltaic array that reflects elements, weather, and environmental factors of a solar cell / array based on machine learning .

According to an embodiment of the present invention, a machine learning based real-time photovoltaic cell or array state predicting method includes: (a) calculating an internal factor, an external factor, and performance data for a plurality of solar arrays each comprising a plurality of solar cells Collecting; (b) machine learning is performed through a predetermined machine learning model to define the relationship between the independent variable and the dependent variable by setting the internal factor and external factor as independent variables and setting the performance data as dependent variables ; (c) calculating predictive performance data for the solar array based on the machine learning, when there are external factors and internal factors collected in real time; And (d) providing the calculated predictive performance data through a user interface of an administrator terminal, wherein the internal factor includes a value for at least one of an internal temperature, a current, a voltage, and a power amount of the solar array Wherein the outlier comprises a value for at least one of an ambient temperature, an irradiation dose, a wind speed, a rainfall, a humidity and a fine dust concentration of the solar array, and the performance data includes a normal operation of the solar cell or the solar array And the like.

According to another embodiment of the present invention, a machine learning based real-time photovoltaic cell or array state prediction server comprises: a memory for storing a machine learning-based real-time photovoltaic cell or a program for performing array state prediction; And a processor for performing the program. Wherein the processor collects internal factors, external factors, and performance data for a plurality of solar arrays each comprising a plurality of solar cells in accordance with the execution of the program, And the performance data is set as a dependent variable to perform a machine learning through a predetermined machine learning model to define a relationship between the independent variable and the dependent variable, The predictive performance data for the solar array is calculated based on the machine learning and the calculated predictive performance data is provided through the user interface of the administrator terminal, A value for at least one of a temperature, a current, a voltage and a power amount, A value of at least one of an ambient temperature, an irradiation amount, a wind speed, a rainfall amount, a humidity, and a fine dust concentration of the ray, and the performance data is data observed as data indicating whether the solar cell or the solar array is operating normally .

The present invention provides an accurate and quick diagnosis result on the state of a solar cell / array based on machine learning and provides a system for predicting malfunction information on an error or the like based thereon, The efficiency can be improved.

As the market of solar power generation system becomes bigger, the demand for monitoring system for maintenance and improvement of efficiency of solar power generation is increasing, and the present invention can be an alternative of this.

Considering that the RPS (New Renewable Energy Mandatory Allotment) system has been introduced in earnest, and it is predicted that the market will be improved due to sound business and professional company forgery after a certain period of time after the initial stage of introduction, Market competitiveness.

1 is a structural view of a solar power plant management system according to an embodiment of the present invention.
2 is a block diagram of a solar array monitoring system in accordance with an embodiment of the present invention.
3 is a reference diagram for explaining the SVM machine learning model.
4 is an example of a manager-side user interface of a solar array monitoring system according to an embodiment of the present invention.
5 is a flowchart illustrating a method of monitoring a solar array according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . Also, when an element is referred to as " comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

In this specification, the term " part " includes a unit realized by hardware, a unit realized by software, and a unit realized by using both. Further, one unit may be implemented using two or more hardware, or two or more units may be implemented by one hardware. On the other hand, 'to' is not limited to software or hardware, 'to' may be configured to be an addressable storage medium, and may be configured to play one or more processors. Thus, by way of example, 'parts' may refer to components such as software components, object-oriented software components, class components and task components, and processes, functions, , Subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. The functions provided in the components and components may be further combined with a smaller number of components and components or further components and components. In addition, the components and components may be implemented to play back one or more CPUs in a device or a secure multimedia card.

The " administrator terminal " referred to below can be implemented as a computer or a portable terminal that can access a server or other terminal through a network. Here, the computer may be, for example, a laptop, a desktop, a laptop, a VR HMD (e.g., HTC VIVE, Oculus Rift, GearVR, DayDream, PSVR etc.) equipped with a web browser . ≪ / RTI > Here, the VR HMD can be used for PC (eg HTC VIVE, Oculus Rift, FOVE, Deepon, etc.), mobile (eg GearVR, DayDream, And stand alone models (e.g., Deepon, PICO, etc.) that are independently implemented. [0002] A portable terminal is a wireless communication device that is guaranteed to be portable and portable, for example, as well as a smart phone, a tablet PC, a wearable device, a Bluetooth (BLE), an NFC, an RFID, , An infrared ray, a WiFi, a LiFi, and the like. The term " network " refers to a connection structure capable of exchanging information between nodes such as terminals and servers. The network includes a local area network (LAN), a wide area network (WAN) (WWW: World Wide Web), wired / wireless data communication network, telephone network, wired / wireless television communication network and the like. Examples of wireless data communication networks include 3G, 4G, 5G, 3GPP, LTE, WIMAX, Wi-Fi, Bluetooth, infrared, Communications, Visible Light Communication (VLC), LiFi, and the like.

Referring to FIG. 1, a solar power plant management system 1 according to an embodiment of the present invention includes a plurality of solar array 100, a solar array 100 monitoring system 200 200) ") and an administrator terminal 300.

The solar power plant management system 1 according to an embodiment of the present invention can predict and detect abnormalities of the solar cell or the solar array 100 based on the monitoring system 200. [

In general, one solar array 100 is composed of a plurality of solar cells. When an anomaly occurs in a solar cell unit, the solar cell array 100 may include other solar cells located in the same row or column as the cell May occur. Which can be diffused due to an abnormal occurrence of the entire solar array 100. The monitoring system 200 predicts an abnormality in a cell unit or an abnormality in an array unit in order to prevent such a case.

The monitoring system 200 collects internal factors and external factors for the solar cell or solar array 100 and performs machine learning on the solar cell or solar array 100 to determine the influence of internal factors and external factors on the solar cell or array And the performance prediction result for the solar cell or array can be provided to the administrator terminal 300 based on internal factors and external factors collected in real time based on the learning results.

The administrator terminal 300 may be configured as a desktop disposed in the controller, but is not limited thereto, and may be configured as a portable terminal carried by the administrator.

Hereinafter, with reference to FIG. 2, a process for deriving a performance prediction result for a solar cell or array based on a machine learning based on the monitoring system 200 will be described in detail.

The monitoring system 200 may include an internal parameter measurement sensor 210, an external parameter measurement sensor 220, a server 240, and an administrator terminal 300.

The internal parameter measurement sensor 210 is a sensor for measuring an internal factor of the solar array 100 with respect to at least one of an internal temperature, a current, a voltage, and a power amount of the solar cell or array. The values for the internal factors may be measured through respective sensors, but may be information that can be collected by the solar array 100 monitoring system 200 itself. That is, electrical components are connected to each solar array 100 to collect power, and such electrical data is an item that can be provided on the user interface in the administrator terminal 300 of the monitoring system 200 , And may be collected on its own without a separate sensor arrangement.

The external factor measurement sensor 220 is a sensor for measuring at least one of ambient temperature, wind speed, solar radiation amount, fine dust concentration, humidity, and rainfall amount of the solar array 100. The ambient temperature above is the temperature related to the ambient space of the solar array 100 and may be the ambient temperature of the array. The external parameter measurement sensor 220 may be provided in a separate sensor form and may be disposed in each solar array 100 or in one area in a space in which the solar array 100 is disposed.

The server 240 may collect such internal factors and external factors from the measurement sensors described above.

Meanwhile, the server 240 may additionally collect the public data 230 from the public entity server 240 and correct the value of the external factor. The public data 230 is data stored in the public entity server 240 (for example, the weather station or the municipality server 240) Is the data relating to the in-values.

The information collected by the external factor measurement sensor 220 is microscopic information collected on the basis of the area where the solar array 100 is disposed and by applying the macroscopic public data 230, Can be supplemented. The external factor measurement sensor 220 may exhibit a very direct property of the solar array 100 although there is a possibility that an error may occur due to an influence of overheating or the like. On the other hand, the public data 230 does not directly indicate the properties of the solar array 100, but has very homogeneous characteristics since it is measured through observation stations disposed at various places in the area. Therefore, when the values collected by the public data 230 and the external factor measurement sensor 220 are combined, a very accurate and complementary value can be obtained as compared with an external factor determined based on only one of them.

The server 240 may perform the correction with the average value of the values collected by the public data 230 and the external parameter measurement sensor 220, but this is merely an example, and the correction may be performed in various ways.

In addition, as a further embodiment, the server 240 may determine the external factors using only the public data 230.

The data compression unit of the server 240 may reduce the dimension of values constituting the internal factors, external factors, and performance data. This is a process performed to increase the speed of big data processing operations.

Here, the performance data refers to data observed by the monitoring system 200 as data indicating the normal operation of the solar cell or the solar array 100. For example, the performance data may be defined as any one of the values 0, 1, and 0.5, where 0 indicates an abnormal state, 1 indicates a normal state, and 0.5 indicates a state requiring confirmation of the state diagnosis.

The data compression section defines training data comprised of internal factors, external factors, and performance data. For example, each training data may comprise values such as internal temperature, irradiation dose, voltage, current, fine dust concentration, performance data, and the like. In this case, it can be defined that each value defines a dimension and the training data consists of n-dimensional values. The data compression unit may reduce the n-dimensional training data to m (less than n) dimensions by using a data compression technique. At this time, PCA (Principal Component Analysis) can be utilized as a data compression technique. PCA is a technique called a principal component analysis technique. It extracts eigenvectors and eigenvalues that show well characteristics of data, extracts only high values of eigenvectors from eigenvectors, and reduces the dimension. It is a technique to minimize. Specifically, the principal component axis that can represent the data well is obtained based on the eigenvector and the eigenvalue, and the data is projected on the principal component axis to reduce the dimension.

Then, the server 240 sets external factors and internal factors as independent variables, sets the performance data as dependent variables, and analyzes the influence between the values constituting the independent variables. For example, the effect of 'solar radiation on voltage value is 60%' can be analyzed. Accordingly, the ratio of the influence of one value in the independent variable to one value of the remaining values can be analyzed. The results of the influence analysis can be used to predict the dependent variable values through the machine learning of future machine learning models, so that more accurate prediction results can be obtained.

The machine learning model generator of the server 240 performs machine learning through a predetermined machine learning model to define the relationship between the independent variable and the dependent variable.

The predetermined machine learning model may be at least one of Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and an ensemble model combining the SVM and ANN.

Preferably, the machine learning model generation unit specifies the order of the machine learning model in advance, and when the error value between the predicted performance data derived by the existing machine learning model and the actual observed performance data exceeds the threshold value, It can be changed to the next machine learning model. Preferably, the machine learning model to be used initially may be an SVM.

As described above, the machine learning model generator sets the internal parameters and the external factors as independent variables, sets the performance data (that is, the normal abnormality of the solar cell or array) as a dependent variable, and corresponds to each of the following machine learning models .

One) SVM (Support Vector Machine)

It is a binary classifier that classifies two categories as a learning model for pattern recognition and data analysis. Based on a given set of data, we create a non-probabilistic binary linear classification model to which category the new data belongs.

The SVM is a binary classification algorithm. When two types of points are given in an N-dimensional place, the SVM creates an (N-1) -dimensional hyperplane and divides these points into two groups. Assuming that there are two types of points that can be linearly separated on paper, the SVM separates these points into two types and finds straight lines that are as far from all points as possible.

The SVM regression model is a functional dependence of the output term {y1, y2, ..., yl} of the input term {x1, x2, ..., xl} (x) can be estimated. Here, the estimation of the function f (x) can result in a problem of minimizing the following risk function. That is, f (x) means an optimal regression function that can define these relationships among multiple input vectors and output terms geometrically. In the SVM model applied in this study, each variable x, y, and f (x) represents a constant demand quantity and dependent variable composed of independent variables, and f (x) represents a predicted constant demand quantity generated through the SVM.

When the SVM is applied to the present invention, the independent variables can be classified into a binary classifier as shown in FIG. For example, assuming that the performance data is 0.5, the right side of the middle thick dotted line area is defined as 0 when the performance data is 0, and the left side is defined as 1 when the performance data is 1, The training data composed of the independent variables are arranged according to the value of the dependent variable (that is, 0, 0.5, or 1). Through such a process, it is possible to perform a machine learning between the independent variable and the dependent variable, and if a new independent variable is input in the future, it is possible to determine which dependent variable is predicted by arranging the independent variable in the SVM space Can be predicted).

2) ANN (Artificial Neural Network)

Just as a natural neural network outputs an output that moves a muscle in response to an input from a sensory organ, an artificial neural network is also a technique that computes an output in response to a given input value. Just as the act of receiving the ball is proficient through training, artificial neural networks function through learning. In general, ANN consists of an input layer, a hidden layer, and an output layer, and each layer is composed of a plurality of nodes, and the nodes between the layers are connected with a constant connection strength. The process of constructing the ANN model is called the learning phase, and the final objective is to determine the model parameters and connection strengths that best fit the given input and output data.

When an ANN is applied to the present invention, it can be derived by mechanically learning a plurality of training data on which weight value a certain independent variable should have in order to be derived as a corresponding dependent variable. Then, when a new independent variable is input, the dependent variable can be predicted by applying the weight value derived from the machine learning.

3) KNN (K-Nearest Neighbors)

The KNN algorithm is a technique for assigning a category using k pieces of data when there is unknown data.

In the kNN algorithm, there are various methods for calculating the distance of the nearest neighbor point. Typically, the Euclidean distance is used.

[Equation 1]

At this time, it is also important to determine a number of neighbor k. Usually, the choice of k depends on the difficulty of the concept to be learned and the number of training data, and is usually set to the square root of the number of training data. It is also possible to set the square root of the number of training data.

If k is set too large, clustering does not work because of the proximity to the surrounding data. If it is too small, it may become noisy data or anomalies and neighbors.

When KNN is applied to the present invention, the training data including the independent variable and the dependent variable is defined as one point on an arbitrary plane and the other remaining training data is defined as one point on the plane. The same training data with one dependent variable is placed close to each other. Using this relationship, when a new independent variable is input, the Euclidean distance formula can be used to predict what the corresponding dependent variable is.

4) Ensemble model

SVM, ANN, and KNN to represent an ensemble model. There are many ways to define an ensemble model.

The main feature of the ensemble concept is the generalization ability of prediction. The general prediction error through the ensemble model is smaller than the general prediction error through a single model (Krogh and Vedelsby, 1995) and has been proven through several studies (Krogh and Sollich, 1997; Naftaly et al., 1997).

The ensemble method is a technique of classifying new data points by weighted vote of predictions after learning algorithms constituting a set of classification criteria. The original ensemble method is Bayesian averaging, but recent algorithms include error-correcting output coding, bagging and boosting.

The data dynamic analysis unit of the server 240 analyzes the pattern of the performance data observed over time in real time, compares the predictive performance data derived by the machine learning model with the predictive performance data based on the pattern, Data can be supplemented.

Specifically, the data dynamic analysis unit can grasp the pattern of performance data according to the time order to analyze the solar cell or array performance data in a time-series manner. At this time, Box-Jenkins' ARIMA analysis methodology used for analysis of univariate time series can be used. The ARIMA methodology is widely used in the economic sector and fisheries resource management, such as the securities market, due to its ability to identify and predict the changes in time series. In general, time series data is composed of trend, cycle, seasonal variation, and irregular fluctuation. The ARIMA analysis serves as a correction prediction model for the machine learning model.

For example, the data dynamic analysis unit can predict new performance data by combining the predictive performance data based on machine learning to zero, but when 0.5 is derived from the time series pattern.

The state diagnosing / predicting unit of the server 240 analyzes the solar array 100 based on at least one machine learning model among the SVM, ANN, KNN, and ensemble models described above in the presence of external factors and internal factors collected in real time. Lt; / RTI > The state diagnostic / prediction unit may compare the predicted performance data with actual observed performance data to determine whether the error value of the predicted performance data is within a preset threshold value. If the threshold value is within a predetermined threshold value, the existing machine learning model is maintained. If the threshold value is exceeded, the control can be changed to another machine learning model as described above.

The state diagnosis / prediction unit of the server 240 may transmit the prediction information to the administrator terminal 300 through the user interface.

The user interface may be a user interface easily visible to the administrator in the administrator terminal 300. The user interface may include a state map that represents the state of each solar cell or each solar array 100 on a color basis. The state map can be expressed, for example, as shown in FIG. That is, one axis (e.g., X axis) of the state map is set as a value of an internal factor, another axis (e.g., Y axis) perpendicular to the one axis is set as a value of an external factor, Each solar cell or each solar array 100 may be placed on the status map according to the state of a factor and external factors, and the state of each cell may be expressed on the basis of color. In other words, cells with high internal temperature and solar radiation of solar cells are displayed at the top left of the graph, and cells with low internal temperature and solar radiation can be displayed at the bottom right of the graph. In this case, similarities in the characteristics of the parameters are arranged, and when the adjacent cells are similar in color to the performance data, it is possible to grasp at a glance the influence of the internal factors or outsiders on the performance data. In FIG. 4, different colors are expressed in different hatching patterns. FIG. 4 is only an example, and the user interface may be implemented in various forms.

Alternatively, cells in one array may be displayed on the screen, and the status of each cell may be displayed by color, so that the presence or absence of a cell may be easily identified in the form of a heat map. Alternatively, if the voltage / current / power amount of a specific cell is significantly different from other cells or significantly different from the usual value, only the information (position information) about the specific cell is informed so that an administrator can easily determine Can be confirmed. In addition, diagnosis / prediction information of the solar cell / array can be provided by a UI that can be easily and quickly understood by the manager in various other ways.

Hereinafter, with reference to FIG. 5, a method of monitoring the solar array 100 according to an embodiment of the present invention will be described in detail.

First, the server 240 collects external factors and internal factors for a plurality of solar array 100 in real time (S110). External factors and internal factors may be collected through the respective sensors or collected by the solar array 100 monitoring system 200 itself or based on the public data 230 of the public entity server 240.

The server 240 reduces the collected data (S120). Specifically, the server 240 constructs training data composed of internal factors, external factors, and performance data, and reduces the dimension of all the training data collected so far, thereby speeding up the processing of the big data .

The server 240 may analyze the influence between the external factor and the internal factor (S130). Specifically, the server 240 sets external factors and internal factors as independent variables, and sets performance data as dependent variables. The server 240 analyzes the influence of any one of the values constituting the independent variable on the other values .

The server 240 selects a machine learning model and performs machine learning (S140). Specifically, the server 240 can select one of the various machine learning models, and derive an algorithm by which a dependent variable can be derived from the independent variable based on the selected machine learning model. Algorithms can be defined in different forms depending on the machine learning model.

The server 240 performs a dynamic analysis of the data to infer the pattern in which the performance data can be seen in a time series flow and to supplement the prediction performance data derived by the machine learning model based on the pattern (S150).

The server 240 may verify the performance data derived by the machine learning model based on the current observation value (i.e., the actual observed performance data), and then change the machine learning model to another model according to the verification result (S160 ).

The server 240 may provide prediction information on the status of the entire solar array 100 or the solar cell to the administrator terminal 300 based on a predetermined user interface.

One embodiment of the present invention may also be embodied in the form of a recording medium including instructions executable by a computer, such as program modules, being executed by a computer. Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. The computer-readable medium may also include computer storage media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data.

While the methods and systems of the present invention have been described in connection with specific embodiments, some or all of those elements or operations may be implemented using a computer system having a general purpose hardware architecture.

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

Claims (14)

A machine learning based real-time photovoltaic cell or array state prediction method, performed by a server,
(a) collecting internal factors, external factors and performance data for a plurality of solar arrays each comprising a plurality of solar cells;
(a-1) reducing the dimension of the values constituting the internal factor, external factor and performance data
(b) machine learning is performed through a predetermined machine learning model to define the relationship between the independent variable and the dependent variable by setting the internal factor and external factor as independent variables and setting the performance data as dependent variables Analyzing the degree of influence between the values constituting the independent variable and analyzing the ratio of the influence of one value in the independent variable to at least one of the other values;
(c) calculating predictive performance data for the solar array based on the machine learning, when there are external factors and internal factors collected in real time; And
(d) analyzing patterns of performance data observed in real time over time, comparing predictive performance data derived by the machine learning model with predictive performance data based on the pattern to correct predictive performance data, and And providing data via a user interface of an administrator terminal,
Wherein the internal factor comprises a value for at least one of an internal temperature, a current, a voltage and an amount of power of the solar array that is directly collected through a monitoring system electrically connected to the solar array, And a value for at least one of an ambient temperature, an irradiation amount, an air speed, a rainfall amount, a humidity and a fine dust concentration of the solar array collected from an external sensor or a server, Data indicating the presence or absence of a normal operation at the present time of the array and being data observed through the monitoring system and indicating identification information regarding an abnormal state,
The step (a)
Receiving public data on external factors stored in the public institution server from a public institution server and correcting data on external factors by combining the public data with external factors collected through the external factor measurement sensor In addition,
The step (a-1)
Based on the principal component analysis technique, the number of internal factors, external factors and performance data composed of n are reduced to a small number,
The step (b)
Wherein the dependent variable derived from the independent variables is a value of the abnormal state that is smaller than the value corresponding to the state diagnosis required state, based on the case where the value of the performance data that is the dependent variable corresponds to the state diagnosis required state, And dividing the independent variables into two groups according to whether the value of the steady state is greater than a value corresponding to a diagnosis required state, thereby mechanically learning the relationship between the independent variable and the dependent variable,
The step (c)
Wherein the order of the plurality of machine learning models is specified in advance and the performance data observed in real time is collected and the predicted performance data is compared with the observed performance data and when the error value of the predicted performance data exceeds a preset threshold And changing the machine learning model to be performed in the step (b) according to the order, wherein the predictive performance data includes at least one of the influence of at least one of the other values in the independent variable The ratio is used in machine learning,
Wherein the user interface comprises:
And a state map indicating the state of each solar cell or each solar array on the basis of color, wherein one axis of the state map is set as an internal factor, and another axis perpendicular to the one axis is set as an outlier Wherein each solar cell or each solar array is placed on the state map according to the state of the internal factor and the external factor.
The method according to claim 1,
The step (a)
Collecting the internal factors through an internal factor measurement sensor of a solar array monitoring system for managing the plurality of solar arrays,
And collecting the external factors through an external factor measurement sensor disposed around the plurality of solar array arrays. delete delete delete delete The method according to claim 1,
The step (b)
Wherein the machine learning is performed based on at least one machine learning model among SVM (Support Vector Machine), K-Nearest Neighbors (KNN), ANN (Artificial Neural Network) Learning based real - time photovoltaic cell or array state prediction method. delete delete delete delete delete A machine learning based real-time photovoltaic cell or array state prediction server,
A memory for storing a program for performing machine learning based real-time photovoltaic cell or array state prediction; And
A processor for performing the program; Lt; / RTI >
The processor, according to the execution of the program,
Collecting internal factors, external factors, and performance data for a plurality of solar arrays, each of which is comprised of a plurality of solar cells,
Reducing the dimensions of the values constituting the internal factor, external factor and performance data,
The machine learning is performed through a predetermined machine learning model to set the internal factor and external factor as independent variables and set the performance data as dependent variables to define a relationship between the independent variable and the dependent variable, Analyzing the degree of influence between the values constituting the independent variable and analyzing the ratio of the influence of any one value in the independent variable to at least one of the other values;
Calculating predictive performance data for the solar array based on the machine learning when external factors and internal factors are collected in real time,
Analyzing a pattern according to time of performance data observed in real time, modifying predictive performance data by comparing predictive performance data derived by the machine learning model with predictive performance data based on the pattern, A user interface of the terminal,
Wherein the internal factor comprises a value for at least one of an internal temperature, a current, a voltage and an amount of power of the solar array that is directly collected through a monitoring system electrically connected to the solar array, And a value for at least one of an ambient temperature, an irradiation amount, a wind speed, a rainfall amount, a humidity and a fine dust concentration of the photovoltaic array collected from an external sensor or a server, Data indicating the presence / absence of normal operation of the optical array, data observed through the monitoring system, and data indicating identification information regarding an abnormal state, a normal state, and a state requiring diagnosis,
The process of collecting the internal factors, external factors,
Receiving public data on external factors stored in the public institution server from a public institution server and correcting data on external factors by combining the public data and external factors collected through the external factor measurement sensor,
The step of reducing the dimension comprises:
Based on the principal component analysis technique, the number of internal factors, external factors and performance data composed of n are reduced to a small number,
Wherein the step of analyzing the ratio of the influence of at least one of the other values in the independent variable,
Wherein the dependent variable derived from the independent variables is a value of the abnormal state that is smaller than the value corresponding to the state diagnosis required state, based on the case where the value of the performance data that is the dependent variable corresponds to the state diagnosis required state, Wherein the controller divides the independent variables into two groups according to whether the value of the steady state is greater than a value corresponding to a diagnosis required state, thereby mechanically learning the relationship between the independent variable and the dependent variable,
Wherein the step of calculating the predictive performance data comprises:
Wherein the order of the plurality of machine learning models is specified in advance and the performance data observed in real time is collected and the predicted performance data is compared with the observed performance data and when the error value of the predicted performance data exceeds a preset threshold According to the above procedure, the machine learning model to be executed is changed,
Wherein the predictive performance data is derived by utilizing a ratio of an influence of one value in the independent variable to at least one of the other values during the machine learning,
Wherein the user interface comprises:
And a state map indicating the state of each solar cell or each solar array on the basis of color, wherein one axis of the state map is set as an internal factor, and another axis perpendicular to the one axis is set as an outlier Wherein each solar cell or each solar array is placed on the state map according to the state of the internal factor and the external factor. A computer-readable recording medium having stored thereon a program for performing a machine learning based real-time photovoltaic cell or array state prediction method according to claim 1.

Images